ETH's anthropomorphic robot hand learns to walk on its own fingers, untethered
ethz · hf · 2026-09-18
ETH Zurich demonstrates an anthropomorphic robotic hand that uses its fingers for locomotion, self-support, and manipulation. Using reinforcement learning with rewards tailored to unequal fingers and a simulator calibrated from hardware, the untethered hand crawls, steers, recovers from falls, types keyboard commands blind, and pushes objects to targets with overhead visual feedback — a compact mobile manipulator with no separate locomotion mechanism.
Related event: ETH Zurich teaches robotic hand to walk on its fingertips(7 posts)→
More from Embodied
- PointZero: a 3D world model pre-trained without robot data hits SOTA on 3D dynamics — YunzhuLiYZ · 2026-09-18
- Dev backports Blender MCP to 3.6.23 for Codex/Astra, generates full 3D character, seeks testers — SpinachOk9137 · 2026-09-18
- Open-Source Dexterous Astra: Zero-Shot Pen Spinning and One-Handed Rubik's Cube in Sim — ZeYanjie · 2026-09-18
- Vacuum with 5 cameras dodges pet poop but can't recognize pee pads — yungcontent · 2026-09-18
- ROSCon Demos Finalized With NVIDIA Thor and RealSense D585/D555 Cameras — chrismatthieu · 2026-09-18
- Training a robotic hand for gestures produced a surprisingly heartwarming moment — techartist_ · 2026-09-18